BigScience BLOOM - 3D Parallelism Explained - Large Language Models - ML Coding Series

BigScience BLOOM - 3D Parallelism Explained - Large Language Models - ML Coding Series

Aleksa Gordić - The AI Epiphany via YouTube Direct link

Intro - focusing on the 3D parallelism!

1 of 18

1 of 18

Intro - focusing on the 3D parallelism!

Class Central Classrooms beta

YouTube playlists curated by Class Central.

Classroom Contents

BigScience BLOOM - 3D Parallelism Explained - Large Language Models - ML Coding Series

Automatically move to the next video in the Classroom when playback concludes

  1. 1 Intro - focusing on the 3D parallelism!
  2. 2 Quick setup
  3. 3 Stepping through the eval script
  4. 4 3D paralellism - model construction
  5. 5 Sharding the embedding table model parallelism
  6. 6 Sharding the transformer layer
  7. 7 LayerNorm fused kernels
  8. 8 Sharding the attention layer
  9. 9 ColumnParallel and RowParallel sharding
  10. 10 Synchronizing input and output embedding tables
  11. 11 Building the dataset data parallelism
  12. 12 3D parallelism - forward pass
  13. 13 Pipeline parallelism communication
  14. 14 Pass through the sharded embedding table
  15. 15 Pass through the sharded transformer layer
  16. 16 Sharded logit and cross-entropy computation
  17. 17 Recap
  18. 18 Outro

Never Stop Learning.

Get personalized course recommendations, track subjects and courses with reminders, and more.

Someone learning on their laptop while sitting on the floor.